Building Safer AGI by introducing Artificial Stupidity. (arXiv:1808.03644v1 [cs.AI])

Artificial Intelligence (AI) achieved super-human performance in a broad
variety of domains. We say that an AI is made Artificially Stupid on a task
when some limitations are deliberately introduced to match a human’s ability to
do the task. An Artificial General Intelligence (AGI) can be made safer by
limiting its computing power and memory, or by introducing Artificial Stupidity
on certain tasks. We survey human intellectual limits and give recommendations
for which limits to implement in order to build a safe AGI.

Source link

Back to top button